reconintra.c 90.6 KB
Newer Older
Jingning Han's avatar
Jingning Han committed
1
/*
Yaowu Xu's avatar
Yaowu Xu committed
2
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
Jingning Han's avatar
Jingning Han committed
3
 *
Yaowu Xu's avatar
Yaowu Xu committed
4
5
6
7
8
9
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
Jingning Han's avatar
Jingning Han committed
10
11
 */

hui su's avatar
hui su committed
12
13
#include <math.h>

Yaowu Xu's avatar
Yaowu Xu committed
14
15
16
#include "./av1_rtcd.h"
#include "./aom_config.h"
#include "./aom_dsp_rtcd.h"
17
#include "aom_ports/system_state.h"
Jingning Han's avatar
Jingning Han committed
18

19
#if CONFIG_HIGHBITDEPTH
Yaowu Xu's avatar
Yaowu Xu committed
20
#include "aom_dsp/aom_dsp_common.h"
21
#endif  // CONFIG_HIGHBITDEPTH
Yaowu Xu's avatar
Yaowu Xu committed
22
#include "aom_mem/aom_mem.h"
23
#include "aom_ports/mem.h"
Yaowu Xu's avatar
Yaowu Xu committed
24
#include "aom_ports/aom_once.h"
25
26
#include "av1/common/reconintra.h"
#include "av1/common/onyxc_int.h"
Jingning Han's avatar
Jingning Han committed
27

28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
enum {
  NEED_LEFT = 1 << 1,
  NEED_ABOVE = 1 << 2,
  NEED_ABOVERIGHT = 1 << 3,
  NEED_ABOVELEFT = 1 << 4,
  NEED_BOTTOMLEFT = 1 << 5,
};

static const uint8_t extend_modes[INTRA_MODES] = {
  NEED_ABOVE | NEED_LEFT,                   // DC
  NEED_ABOVE,                               // V
  NEED_LEFT,                                // H
  NEED_ABOVE | NEED_ABOVERIGHT,             // D45
  NEED_LEFT | NEED_ABOVE | NEED_ABOVELEFT,  // D135
  NEED_LEFT | NEED_ABOVE | NEED_ABOVELEFT,  // D117
  NEED_LEFT | NEED_ABOVE | NEED_ABOVELEFT,  // D153
  NEED_LEFT | NEED_BOTTOMLEFT,              // D207
  NEED_ABOVE | NEED_ABOVERIGHT,             // D63
46
47
48
#if CONFIG_ALT_INTRA
  NEED_LEFT | NEED_ABOVE,                   // SMOOTH
#endif                                      // CONFIG_ALT_INTRA
49
50
  NEED_LEFT | NEED_ABOVE | NEED_ABOVELEFT,  // TM
};
Jingning Han's avatar
Jingning Han committed
51

52
53
54
55
static const uint16_t orders_128x128[1] = { 0 };
static const uint16_t orders_128x64[2] = { 0, 1 };
static const uint16_t orders_64x128[2] = { 0, 1 };
static const uint16_t orders_64x64[4] = {
clang-format's avatar
clang-format committed
56
  0, 1, 2, 3,
57
};
58
static const uint16_t orders_64x32[8] = {
clang-format's avatar
clang-format committed
59
  0, 2, 1, 3, 4, 6, 5, 7,
60
};
61
static const uint16_t orders_32x64[8] = {
clang-format's avatar
clang-format committed
62
  0, 1, 2, 3, 4, 5, 6, 7,
63
};
64
static const uint16_t orders_32x32[16] = {
clang-format's avatar
clang-format committed
65
  0, 1, 4, 5, 2, 3, 6, 7, 8, 9, 12, 13, 10, 11, 14, 15,
66
};
67
static const uint16_t orders_32x16[32] = {
clang-format's avatar
clang-format committed
68
69
  0,  2,  8,  10, 1,  3,  9,  11, 4,  6,  12, 14, 5,  7,  13, 15,
  16, 18, 24, 26, 17, 19, 25, 27, 20, 22, 28, 30, 21, 23, 29, 31,
70
};
71
static const uint16_t orders_16x32[32] = {
clang-format's avatar
clang-format committed
72
73
  0,  1,  2,  3,  8,  9,  10, 11, 4,  5,  6,  7,  12, 13, 14, 15,
  16, 17, 18, 19, 24, 25, 26, 27, 20, 21, 22, 23, 28, 29, 30, 31,
74
};
75
static const uint16_t orders_16x16[64] = {
clang-format's avatar
clang-format committed
76
77
78
79
  0,  1,  4,  5,  16, 17, 20, 21, 2,  3,  6,  7,  18, 19, 22, 23,
  8,  9,  12, 13, 24, 25, 28, 29, 10, 11, 14, 15, 26, 27, 30, 31,
  32, 33, 36, 37, 48, 49, 52, 53, 34, 35, 38, 39, 50, 51, 54, 55,
  40, 41, 44, 45, 56, 57, 60, 61, 42, 43, 46, 47, 58, 59, 62, 63,
80
};
81

82
#if CONFIG_CB4X4 || CONFIG_EXT_PARTITION
83
static const uint16_t orders_16x8[128] = {
clang-format's avatar
clang-format committed
84
85
86
87
88
89
90
91
  0,  2,  8,  10, 32,  34,  40,  42,  1,  3,  9,  11, 33,  35,  41,  43,
  4,  6,  12, 14, 36,  38,  44,  46,  5,  7,  13, 15, 37,  39,  45,  47,
  16, 18, 24, 26, 48,  50,  56,  58,  17, 19, 25, 27, 49,  51,  57,  59,
  20, 22, 28, 30, 52,  54,  60,  62,  21, 23, 29, 31, 53,  55,  61,  63,
  64, 66, 72, 74, 96,  98,  104, 106, 65, 67, 73, 75, 97,  99,  105, 107,
  68, 70, 76, 78, 100, 102, 108, 110, 69, 71, 77, 79, 101, 103, 109, 111,
  80, 82, 88, 90, 112, 114, 120, 122, 81, 83, 89, 91, 113, 115, 121, 123,
  84, 86, 92, 94, 116, 118, 124, 126, 85, 87, 93, 95, 117, 119, 125, 127,
92
};
93
static const uint16_t orders_8x16[128] = {
clang-format's avatar
clang-format committed
94
95
96
97
98
  0,  1,  2,  3,  8,  9,  10, 11, 32,  33,  34,  35,  40,  41,  42,  43,
  4,  5,  6,  7,  12, 13, 14, 15, 36,  37,  38,  39,  44,  45,  46,  47,
  16, 17, 18, 19, 24, 25, 26, 27, 48,  49,  50,  51,  56,  57,  58,  59,
  20, 21, 22, 23, 28, 29, 30, 31, 52,  53,  54,  55,  60,  61,  62,  63,
  64, 65, 66, 67, 72, 73, 74, 75, 96,  97,  98,  99,  104, 105, 106, 107,
99
100
101
102
  68, 69, 70, 71, 76, 77, 78, 79, 100, 101, 102, 103, 108, 109, 110, 111,
  80, 81, 82, 83, 88, 89, 90, 91, 112, 113, 114, 115, 120, 121, 122, 123,
  84, 85, 86, 87, 92, 93, 94, 95, 116, 117, 118, 119, 124, 125, 126, 127,
};
103
static const uint16_t orders_8x8[256] = {
clang-format's avatar
clang-format committed
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
  0,   1,   4,   5,   16,  17,  20,  21,  64,  65,  68,  69,  80,  81,  84,
  85,  2,   3,   6,   7,   18,  19,  22,  23,  66,  67,  70,  71,  82,  83,
  86,  87,  8,   9,   12,  13,  24,  25,  28,  29,  72,  73,  76,  77,  88,
  89,  92,  93,  10,  11,  14,  15,  26,  27,  30,  31,  74,  75,  78,  79,
  90,  91,  94,  95,  32,  33,  36,  37,  48,  49,  52,  53,  96,  97,  100,
  101, 112, 113, 116, 117, 34,  35,  38,  39,  50,  51,  54,  55,  98,  99,
  102, 103, 114, 115, 118, 119, 40,  41,  44,  45,  56,  57,  60,  61,  104,
  105, 108, 109, 120, 121, 124, 125, 42,  43,  46,  47,  58,  59,  62,  63,
  106, 107, 110, 111, 122, 123, 126, 127, 128, 129, 132, 133, 144, 145, 148,
  149, 192, 193, 196, 197, 208, 209, 212, 213, 130, 131, 134, 135, 146, 147,
  150, 151, 194, 195, 198, 199, 210, 211, 214, 215, 136, 137, 140, 141, 152,
  153, 156, 157, 200, 201, 204, 205, 216, 217, 220, 221, 138, 139, 142, 143,
  154, 155, 158, 159, 202, 203, 206, 207, 218, 219, 222, 223, 160, 161, 164,
  165, 176, 177, 180, 181, 224, 225, 228, 229, 240, 241, 244, 245, 162, 163,
  166, 167, 178, 179, 182, 183, 226, 227, 230, 231, 242, 243, 246, 247, 168,
  169, 172, 173, 184, 185, 188, 189, 232, 233, 236, 237, 248, 249, 252, 253,
  170, 171, 174, 175, 186, 187, 190, 191, 234, 235, 238, 239, 250, 251, 254,
  255,
122
};
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282

#if CONFIG_CB4X4 && CONFIG_EXT_PARTITION
static const uint16_t orders_4x8[512] = {
  0,   1,   2,   3,   8,   9,   10,  11,  32,  33,  34,  35,  40,  41,  42,
  43,  128, 129, 130, 131, 136, 137, 138, 139, 160, 161, 162, 163, 168, 169,
  170, 171, 4,   5,   6,   7,   12,  13,  14,  15,  36,  37,  38,  39,  44,
  45,  46,  47,  132, 133, 134, 135, 140, 141, 142, 143, 164, 165, 166, 167,
  172, 173, 174, 175, 16,  17,  18,  19,  24,  25,  26,  27,  48,  49,  50,
  51,  56,  57,  58,  59,  144, 145, 146, 147, 152, 153, 154, 155, 176, 177,
  178, 179, 184, 185, 186, 187, 20,  21,  22,  23,  28,  29,  30,  31,  52,
  53,  54,  55,  60,  61,  62,  63,  148, 149, 150, 151, 156, 157, 158, 159,
  180, 181, 182, 183, 188, 189, 190, 191, 64,  65,  66,  67,  72,  73,  74,
  75,  96,  97,  98,  99,  104, 105, 106, 107, 192, 193, 194, 195, 200, 201,
  202, 203, 224, 225, 226, 227, 232, 233, 234, 235, 68,  69,  70,  71,  76,
  77,  78,  79,  100, 101, 102, 103, 108, 109, 110, 111, 196, 197, 198, 199,
  204, 205, 206, 207, 228, 229, 230, 231, 236, 237, 238, 239, 80,  81,  82,
  83,  88,  89,  90,  91,  112, 113, 114, 115, 120, 121, 122, 123, 208, 209,
  210, 211, 216, 217, 218, 219, 240, 241, 242, 243, 248, 249, 250, 251, 84,
  85,  86,  87,  92,  93,  94,  95,  116, 117, 118, 119, 124, 125, 126, 127,
  212, 213, 214, 215, 220, 221, 222, 223, 244, 245, 246, 247, 252, 253, 254,
  255, 256, 257, 258, 259, 264, 265, 266, 267, 288, 289, 290, 291, 296, 297,
  298, 299, 384, 385, 386, 387, 392, 393, 394, 395, 416, 417, 418, 419, 424,
  425, 426, 427, 260, 261, 262, 263, 268, 269, 270, 271, 292, 293, 294, 295,
  300, 301, 302, 303, 388, 389, 390, 391, 396, 397, 398, 399, 420, 421, 422,
  423, 428, 429, 430, 431, 272, 273, 274, 275, 280, 281, 282, 283, 304, 305,
  306, 307, 312, 313, 314, 315, 400, 401, 402, 403, 408, 409, 410, 411, 432,
  433, 434, 435, 440, 441, 442, 443, 276, 277, 278, 279, 284, 285, 286, 287,
  308, 309, 310, 311, 316, 317, 318, 319, 404, 405, 406, 407, 412, 413, 414,
  415, 436, 437, 438, 439, 444, 445, 446, 447, 320, 321, 322, 323, 328, 329,
  330, 331, 352, 353, 354, 355, 360, 361, 362, 363, 448, 449, 450, 451, 456,
  457, 458, 459, 480, 481, 482, 483, 488, 489, 490, 491, 324, 325, 326, 327,
  332, 333, 334, 335, 356, 357, 358, 359, 364, 365, 366, 367, 452, 453, 454,
  455, 460, 461, 462, 463, 484, 485, 486, 487, 492, 493, 494, 495, 336, 337,
  338, 339, 344, 345, 346, 347, 368, 369, 370, 371, 376, 377, 378, 379, 464,
  465, 466, 467, 472, 473, 474, 475, 496, 497, 498, 499, 504, 505, 506, 507,
  340, 341, 342, 343, 348, 349, 350, 351, 372, 373, 374, 375, 380, 381, 382,
  383, 468, 469, 470, 471, 476, 477, 478, 479, 500, 501, 502, 503, 508, 509,
  510, 511,
};

static const uint16_t orders_8x4[512] = {
  0,   2,   8,   10,  32,  34,  40,  42,  128, 130, 136, 138, 160, 162, 168,
  170, 1,   3,   9,   11,  33,  35,  41,  43,  129, 131, 137, 139, 161, 163,
  169, 171, 4,   6,   12,  14,  36,  38,  44,  46,  132, 134, 140, 142, 164,
  166, 172, 174, 5,   7,   13,  15,  37,  39,  45,  47,  133, 135, 141, 143,
  165, 167, 173, 175, 16,  18,  24,  26,  48,  50,  56,  58,  144, 146, 152,
  154, 176, 178, 184, 186, 17,  19,  25,  27,  49,  51,  57,  59,  145, 147,
  153, 155, 177, 179, 185, 187, 20,  22,  28,  30,  52,  54,  60,  62,  148,
  150, 156, 158, 180, 182, 188, 190, 21,  23,  29,  31,  53,  55,  61,  63,
  149, 151, 157, 159, 181, 183, 189, 191, 64,  66,  72,  74,  96,  98,  104,
  106, 192, 194, 200, 202, 224, 226, 232, 234, 65,  67,  73,  75,  97,  99,
  105, 107, 193, 195, 201, 203, 225, 227, 233, 235, 68,  70,  76,  78,  100,
  102, 108, 110, 196, 198, 204, 206, 228, 230, 236, 238, 69,  71,  77,  79,
  101, 103, 109, 111, 197, 199, 205, 207, 229, 231, 237, 239, 80,  82,  88,
  90,  112, 114, 120, 122, 208, 210, 216, 218, 240, 242, 248, 250, 81,  83,
  89,  91,  113, 115, 121, 123, 209, 211, 217, 219, 241, 243, 249, 251, 84,
  86,  92,  94,  116, 118, 124, 126, 212, 214, 220, 222, 244, 246, 252, 254,
  85,  87,  93,  95,  117, 119, 125, 127, 213, 215, 221, 223, 245, 247, 253,
  255, 256, 258, 264, 266, 288, 290, 296, 298, 384, 386, 392, 394, 416, 418,
  424, 426, 257, 259, 265, 267, 289, 291, 297, 299, 385, 387, 393, 395, 417,
  419, 425, 427, 260, 262, 268, 270, 292, 294, 300, 302, 388, 390, 396, 398,
  420, 422, 428, 430, 261, 263, 269, 271, 293, 295, 301, 303, 389, 391, 397,
  399, 421, 423, 429, 431, 272, 274, 280, 282, 304, 306, 312, 314, 400, 402,
  408, 410, 432, 434, 440, 442, 273, 275, 281, 283, 305, 307, 313, 315, 401,
  403, 409, 411, 433, 435, 441, 443, 276, 278, 284, 286, 308, 310, 316, 318,
  404, 406, 412, 414, 436, 438, 444, 446, 277, 279, 285, 287, 309, 311, 317,
  319, 405, 407, 413, 415, 437, 439, 445, 447, 320, 322, 328, 330, 352, 354,
  360, 362, 448, 450, 456, 458, 480, 482, 488, 490, 321, 323, 329, 331, 353,
  355, 361, 363, 449, 451, 457, 459, 481, 483, 489, 491, 324, 326, 332, 334,
  356, 358, 364, 366, 452, 454, 460, 462, 484, 486, 492, 494, 325, 327, 333,
  335, 357, 359, 365, 367, 453, 455, 461, 463, 485, 487, 493, 495, 336, 338,
  344, 346, 368, 370, 376, 378, 464, 466, 472, 474, 496, 498, 504, 506, 337,
  339, 345, 347, 369, 371, 377, 379, 465, 467, 473, 475, 497, 499, 505, 507,
  340, 342, 348, 350, 372, 374, 380, 382, 468, 470, 476, 478, 500, 502, 508,
  510, 341, 343, 349, 351, 373, 375, 381, 383, 469, 471, 477, 479, 501, 503,
  509, 511,
};

static const uint16_t orders_4x4[1024] = {
  0,    1,    4,    5,    16,   17,   20,   21,   64,   65,   68,   69,   80,
  81,   84,   85,   256,  257,  260,  261,  272,  273,  276,  277,  320,  321,
  324,  325,  336,  337,  340,  341,  2,    3,    6,    7,    18,   19,   22,
  23,   66,   67,   70,   71,   82,   83,   86,   87,   258,  259,  262,  263,
  274,  275,  278,  279,  322,  323,  326,  327,  338,  339,  342,  343,  8,
  9,    12,   13,   24,   25,   28,   29,   72,   73,   76,   77,   88,   89,
  92,   93,   264,  265,  268,  269,  280,  281,  284,  285,  328,  329,  332,
  333,  344,  345,  348,  349,  10,   11,   14,   15,   26,   27,   30,   31,
  74,   75,   78,   79,   90,   91,   94,   95,   266,  267,  270,  271,  282,
  283,  286,  287,  330,  331,  334,  335,  346,  347,  350,  351,  32,   33,
  36,   37,   48,   49,   52,   53,   96,   97,   100,  101,  112,  113,  116,
  117,  288,  289,  292,  293,  304,  305,  308,  309,  352,  353,  356,  357,
  368,  369,  372,  373,  34,   35,   38,   39,   50,   51,   54,   55,   98,
  99,   102,  103,  114,  115,  118,  119,  290,  291,  294,  295,  306,  307,
  310,  311,  354,  355,  358,  359,  370,  371,  374,  375,  40,   41,   44,
  45,   56,   57,   60,   61,   104,  105,  108,  109,  120,  121,  124,  125,
  296,  297,  300,  301,  312,  313,  316,  317,  360,  361,  364,  365,  376,
  377,  380,  381,  42,   43,   46,   47,   58,   59,   62,   63,   106,  107,
  110,  111,  122,  123,  126,  127,  298,  299,  302,  303,  314,  315,  318,
  319,  362,  363,  366,  367,  378,  379,  382,  383,  128,  129,  132,  133,
  144,  145,  148,  149,  192,  193,  196,  197,  208,  209,  212,  213,  384,
  385,  388,  389,  400,  401,  404,  405,  448,  449,  452,  453,  464,  465,
  468,  469,  130,  131,  134,  135,  146,  147,  150,  151,  194,  195,  198,
  199,  210,  211,  214,  215,  386,  387,  390,  391,  402,  403,  406,  407,
  450,  451,  454,  455,  466,  467,  470,  471,  136,  137,  140,  141,  152,
  153,  156,  157,  200,  201,  204,  205,  216,  217,  220,  221,  392,  393,
  396,  397,  408,  409,  412,  413,  456,  457,  460,  461,  472,  473,  476,
  477,  138,  139,  142,  143,  154,  155,  158,  159,  202,  203,  206,  207,
  218,  219,  222,  223,  394,  395,  398,  399,  410,  411,  414,  415,  458,
  459,  462,  463,  474,  475,  478,  479,  160,  161,  164,  165,  176,  177,
  180,  181,  224,  225,  228,  229,  240,  241,  244,  245,  416,  417,  420,
  421,  432,  433,  436,  437,  480,  481,  484,  485,  496,  497,  500,  501,
  162,  163,  166,  167,  178,  179,  182,  183,  226,  227,  230,  231,  242,
  243,  246,  247,  418,  419,  422,  423,  434,  435,  438,  439,  482,  483,
  486,  487,  498,  499,  502,  503,  168,  169,  172,  173,  184,  185,  188,
  189,  232,  233,  236,  237,  248,  249,  252,  253,  424,  425,  428,  429,
  440,  441,  444,  445,  488,  489,  492,  493,  504,  505,  508,  509,  170,
  171,  174,  175,  186,  187,  190,  191,  234,  235,  238,  239,  250,  251,
  254,  255,  426,  427,  430,  431,  442,  443,  446,  447,  490,  491,  494,
  495,  506,  507,  510,  511,  512,  513,  516,  517,  528,  529,  532,  533,
  576,  577,  580,  581,  592,  593,  596,  597,  768,  769,  772,  773,  784,
  785,  788,  789,  832,  833,  836,  837,  848,  849,  852,  853,  514,  515,
  518,  519,  530,  531,  534,  535,  578,  579,  582,  583,  594,  595,  598,
  599,  770,  771,  774,  775,  786,  787,  790,  791,  834,  835,  838,  839,
  850,  851,  854,  855,  520,  521,  524,  525,  536,  537,  540,  541,  584,
  585,  588,  589,  600,  601,  604,  605,  776,  777,  780,  781,  792,  793,
  796,  797,  840,  841,  844,  845,  856,  857,  860,  861,  522,  523,  526,
  527,  538,  539,  542,  543,  586,  587,  590,  591,  602,  603,  606,  607,
  778,  779,  782,  783,  794,  795,  798,  799,  842,  843,  846,  847,  858,
  859,  862,  863,  544,  545,  548,  549,  560,  561,  564,  565,  608,  609,
  612,  613,  624,  625,  628,  629,  800,  801,  804,  805,  816,  817,  820,
  821,  864,  865,  868,  869,  880,  881,  884,  885,  546,  547,  550,  551,
  562,  563,  566,  567,  610,  611,  614,  615,  626,  627,  630,  631,  802,
  803,  806,  807,  818,  819,  822,  823,  866,  867,  870,  871,  882,  883,
  886,  887,  552,  553,  556,  557,  568,  569,  572,  573,  616,  617,  620,
  621,  632,  633,  636,  637,  808,  809,  812,  813,  824,  825,  828,  829,
  872,  873,  876,  877,  888,  889,  892,  893,  554,  555,  558,  559,  570,
  571,  574,  575,  618,  619,  622,  623,  634,  635,  638,  639,  810,  811,
  814,  815,  826,  827,  830,  831,  874,  875,  878,  879,  890,  891,  894,
  895,  640,  641,  644,  645,  656,  657,  660,  661,  704,  705,  708,  709,
  720,  721,  724,  725,  896,  897,  900,  901,  912,  913,  916,  917,  960,
  961,  964,  965,  976,  977,  980,  981,  642,  643,  646,  647,  658,  659,
  662,  663,  706,  707,  710,  711,  722,  723,  726,  727,  898,  899,  902,
  903,  914,  915,  918,  919,  962,  963,  966,  967,  978,  979,  982,  983,
  648,  649,  652,  653,  664,  665,  668,  669,  712,  713,  716,  717,  728,
  729,  732,  733,  904,  905,  908,  909,  920,  921,  924,  925,  968,  969,
  972,  973,  984,  985,  988,  989,  650,  651,  654,  655,  666,  667,  670,
  671,  714,  715,  718,  719,  730,  731,  734,  735,  906,  907,  910,  911,
  922,  923,  926,  927,  970,  971,  974,  975,  986,  987,  990,  991,  672,
  673,  676,  677,  688,  689,  692,  693,  736,  737,  740,  741,  752,  753,
  756,  757,  928,  929,  932,  933,  944,  945,  948,  949,  992,  993,  996,
  997,  1008, 1009, 1012, 1013, 674,  675,  678,  679,  690,  691,  694,  695,
  738,  739,  742,  743,  754,  755,  758,  759,  930,  931,  934,  935,  946,
  947,  950,  951,  994,  995,  998,  999,  1010, 1011, 1014, 1015, 680,  681,
  684,  685,  696,  697,  700,  701,  744,  745,  748,  749,  760,  761,  764,
  765,  936,  937,  940,  941,  952,  953,  956,  957,  1000, 1001, 1004, 1005,
  1016, 1017, 1020, 1021, 682,  683,  686,  687,  698,  699,  702,  703,  746,
  747,  750,  751,  762,  763,  766,  767,  938,  939,  942,  943,  954,  955,
  958,  959,  1002, 1003, 1006, 1007, 1018, 1019, 1022, 1023,
};
#endif
283
#endif  // CONFIG_CB4X4 || CONFIG_EXT_PARTITION
284

285
#if CONFIG_EXT_PARTITION
clang-format's avatar
clang-format committed
286
/* clang-format off */
287
static const uint16_t *const orders[BLOCK_SIZES] = {
Jingning Han's avatar
Jingning Han committed
288
289
#if CONFIG_CB4X4
  // 2X2,         2X4,            4X2
290
291
292
293
294
295
  orders_4x4,     orders_4x4,     orders_4x4,
  //                              4X4
                                  orders_4x4,
  // 4X8,         8X4,            8X8
  orders_4x8,     orders_8x4,     orders_8x8,
#else
296
297
298
299
  //                              4X4
                                  orders_8x8,
  // 4X8,         8X4,            8X8
  orders_8x8,     orders_8x8,     orders_8x8,
300
#endif
301
302
303
304
305
306
307
308
  // 8X16,        16X8,           16X16
  orders_8x16,    orders_16x8,    orders_16x16,
  // 16X32,       32X16,          32X32
  orders_16x32,   orders_32x16,   orders_32x32,
  // 32X64,       64X32,          64X64
  orders_32x64,   orders_64x32,   orders_64x64,
  // 64x128,      128x64,         128x128
  orders_64x128,  orders_128x64,  orders_128x128
309
};
clang-format's avatar
clang-format committed
310
/* clang-format on */
311
#else
clang-format's avatar
clang-format committed
312
/* clang-format off */
313
static const uint16_t *const orders[BLOCK_SIZES] = {
Jingning Han's avatar
Jingning Han committed
314
315
#if CONFIG_CB4X4
  // 2X2,         2X4,            4X2
316
317
318
319
320
321
  orders_8x8,     orders_8x8,     orders_8x8,
  //                              4X4
                                  orders_8x8,
  // 4X8,         8X4,            8X8
  orders_8x16,    orders_16x8,    orders_16x16,
#else
322
323
324
325
  //                              4X4
                                  orders_16x16,
  // 4X8,         8X4,            8X8
  orders_16x16,   orders_16x16,   orders_16x16,
326
#endif
327
328
329
330
331
332
333
  // 8X16,        16X8,           16X16
  orders_16x32,   orders_32x16,   orders_32x32,
  // 16X32,       32X16,          32X32
  orders_32x64,   orders_64x32,   orders_64x64,
  // 32X64,       64X32,          64X64
  orders_64x128,  orders_128x64,  orders_128x128
};
clang-format's avatar
clang-format committed
334
/* clang-format on */
335
336
#endif  // CONFIG_EXT_PARTITION

337
#if CONFIG_EXT_PARTITION_TYPES
338
static const uint16_t orders_verta_64x64[4] = {
clang-format's avatar
clang-format committed
339
  0, 2, 1, 2,
340
};
341
static const uint16_t orders_verta_32x32[16] = {
clang-format's avatar
clang-format committed
342
  0, 2, 4, 6, 1, 2, 5, 6, 8, 10, 12, 14, 9, 10, 13, 14,
343
};
344
static const uint16_t orders_verta_16x16[64] = {
clang-format's avatar
clang-format committed
345
346
347
348
  0,  2,  4,  6,  16, 18, 20, 22, 1,  2,  5,  6,  17, 18, 21, 22,
  8,  10, 12, 14, 24, 26, 28, 30, 9,  10, 13, 14, 25, 26, 29, 30,
  32, 34, 36, 38, 48, 50, 52, 54, 33, 34, 37, 38, 49, 50, 53, 54,
  40, 42, 44, 46, 56, 58, 60, 62, 41, 42, 45, 46, 57, 58, 61, 62,
349
};
350
#if CONFIG_EXT_PARTITION || CONFIG_CB4X4
351
static const uint16_t orders_verta_8x8[256] = {
clang-format's avatar
clang-format committed
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
  0,   2,   4,   6,   16,  18,  20,  22,  64,  66,  68,  70,  80,  82,  84,
  86,  1,   2,   5,   6,   17,  18,  21,  22,  65,  66,  69,  70,  81,  82,
  85,  86,  8,   10,  12,  14,  24,  26,  28,  30,  72,  74,  76,  78,  88,
  90,  92,  94,  9,   10,  13,  14,  25,  26,  29,  30,  73,  74,  77,  78,
  89,  90,  93,  94,  32,  34,  36,  38,  48,  50,  52,  54,  96,  98,  100,
  102, 112, 114, 116, 118, 33,  34,  37,  38,  49,  50,  53,  54,  97,  98,
  101, 102, 113, 114, 117, 118, 40,  42,  44,  46,  56,  58,  60,  62,  104,
  106, 108, 110, 120, 122, 124, 126, 41,  42,  45,  46,  57,  58,  61,  62,
  105, 106, 109, 110, 121, 122, 125, 126, 128, 130, 132, 134, 144, 146, 148,
  150, 192, 194, 196, 198, 208, 210, 212, 214, 129, 130, 133, 134, 145, 146,
  149, 150, 193, 194, 197, 198, 209, 210, 213, 214, 136, 138, 140, 142, 152,
  154, 156, 158, 200, 202, 204, 206, 216, 218, 220, 222, 137, 138, 141, 142,
  153, 154, 157, 158, 201, 202, 205, 206, 217, 218, 221, 222, 160, 162, 164,
  166, 176, 178, 180, 182, 224, 226, 228, 230, 240, 242, 244, 246, 161, 162,
  165, 166, 177, 178, 181, 182, 225, 226, 229, 230, 241, 242, 245, 246, 168,
  170, 172, 174, 184, 186, 188, 190, 232, 234, 236, 238, 248, 250, 252, 254,
  169, 170, 173, 174, 185, 186, 189, 190, 233, 234, 237, 238, 249, 250, 253,
  254,
370
};
371
#endif  // CONFIG_EXT_PARTITION || CONFIG_CB4X4
clang-format's avatar
clang-format committed
372

373
#if CONFIG_EXT_PARTITION
clang-format's avatar
clang-format committed
374
/* clang-format off */
375
static const uint16_t *const orders_verta[BLOCK_SIZES] = {
376
377
378
379
#if CONFIG_CB4X4
  // 2X2,           2X4,              4X2
  orders_4x4,       orders_4x4,       orders_4x4,
#endif
380
381
382
383
384
385
386
387
388
389
390
391
392
  //                                  4X4
                                      orders_verta_8x8,
  // 4X8,           8X4,              8X8
  orders_verta_8x8, orders_verta_8x8, orders_verta_8x8,
  // 8X16,          16X8,             16X16
  orders_8x16,      orders_16x8,      orders_verta_16x16,
  // 16X32,         32X16,            32X32
  orders_16x32,     orders_32x16,     orders_verta_32x32,
  // 32X64,         64X32,            64X64
  orders_32x64,     orders_64x32,     orders_verta_64x64,
  // 64x128,        128x64,           128x128
  orders_64x128,    orders_128x64,    orders_128x128
};
clang-format's avatar
clang-format committed
393
/* clang-format on */
394
#else
clang-format's avatar
clang-format committed
395
/* clang-format off */
396
static const uint16_t *const orders_verta[BLOCK_SIZES] = {
397
398
399
400
401
402
403
404
#if CONFIG_CB4X4
  // 2X2,             2X4,                4X2
  orders_verta_8x8,   orders_verta_8x8,   orders_verta_8x8,
  //                                      4X4
                                          orders_verta_8x8,
  // 4X8,             8X4,                8X8
  orders_verta_8x8,   orders_verta_8x8,   orders_verta_16x16,
#else
405
406
407
408
  //                                      4X4
                                          orders_verta_16x16,
  // 4X8,             8X4,                8X8
  orders_verta_16x16, orders_verta_16x16, orders_verta_16x16,
409
#endif
410
411
412
413
414
415
  // 8X16,            16X8,               16X16
  orders_16x32,       orders_32x16,       orders_verta_32x32,
  // 16X32,           32X16,              32X32
  orders_32x64,       orders_64x32,       orders_verta_64x64,
  // 32X64,           64X32,              64X64
  orders_64x128,      orders_128x64,      orders_128x128
416
};
clang-format's avatar
clang-format committed
417
/* clang-format on */
418
#endif  // CONFIG_EXT_PARTITION
419
#endif  // CONFIG_EXT_PARTITION_TYPES
420

421
422
static int has_top_right(BLOCK_SIZE bsize, int mi_row, int mi_col,
                         int top_available, int right_available,
423
#if CONFIG_EXT_PARTITION_TYPES
Yaowu Xu's avatar
Yaowu Xu committed
424
                         PARTITION_TYPE partition,
425
#endif
426
427
428
                         TX_SIZE txsz, int row_off, int col_off, int ss_x) {
  if (!top_available || !right_available) return 0;

429
#if !CONFIG_CB4X4
430
431
432
433
434
435
  // TODO(bshacklett, huisu): Currently the RD loop traverses 4X8 blocks in
  // inverted N order while in the bitstream the subblocks are stored in Z
  // order. This discrepancy makes this function incorrect when considering 4X8
  // blocks in the RD loop, so we disable the extended right edge for these
  // blocks. The correct solution is to change the bitstream to store these
  // blocks in inverted N order, and then update this function appropriately.
436
  if (bsize == BLOCK_4X8 && row_off == 1) return 0;
437
#endif
438

439
440
441
  const int bw_unit = block_size_wide[bsize] >> tx_size_wide_log2[0];
  const int plane_bw_unit = AOMMAX(bw_unit >> ss_x, 1);
  const int top_right_count_unit = tx_size_wide_unit[txsz];
442

443
444
  // Special handling for block sizes 4x8 and 4x4.
  if (ss_x == 0 && bw_unit < 2 && col_off == 0) return 1;
445

446
447
448
449
450
  if (row_off > 0) {  // Just need to check if enough pixels on the right.
    return col_off + top_right_count_unit < plane_bw_unit;
  } else {
    // All top-right pixels are in the block above, which is already available.
    if (col_off + top_right_count_unit < plane_bw_unit) return 1;
451

452
453
454
455
    const int bw_in_mi_log2 = mi_width_log2_lookup[bsize];
    const int bh_in_mi_log2 = mi_height_log2_lookup[bsize];
    const int blk_row_in_sb = (mi_row & MAX_MIB_MASK) >> bh_in_mi_log2;
    const int blk_col_in_sb = (mi_col & MAX_MIB_MASK) >> bw_in_mi_log2;
456

457
458
459
    // Top row of superblock: so top-right pixels are in the top and/or
    // top-right superblocks, both of which are already available.
    if (blk_row_in_sb == 0) return 1;
460

461
462
463
    // Rightmost column of superblock (and not the top row): so top-right pixels
    // fall in the right superblock, which is not available yet.
    if (((blk_col_in_sb + 1) << bw_in_mi_log2) >= MAX_MIB_SIZE) return 0;
464

465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
    // General case (neither top row nor rightmost column): check if the
    // top-right block is coded before the current block.
    const uint16_t *const order =
#if CONFIG_EXT_PARTITION_TYPES
        (partition == PARTITION_VERT_A) ? orders_verta[bsize] :
#endif  // CONFIG_EXT_PARTITION_TYPES
                                        orders[bsize];
    const int this_blk_index =
        ((blk_row_in_sb + 0) << (MAX_MIB_SIZE_LOG2 - bw_in_mi_log2)) +
        blk_col_in_sb + 0;
    const uint16_t this_blk_order = order[this_blk_index];
    const int tr_blk_index =
        ((blk_row_in_sb - 1) << (MAX_MIB_SIZE_LOG2 - bw_in_mi_log2)) +
        blk_col_in_sb + 1;
    const uint16_t tr_blk_order = order[tr_blk_index];
    return tr_blk_order < this_blk_order;
481
482
483
  }
}

484
485
486
487
488
489
490
static int has_bottom_left(BLOCK_SIZE bsize, int mi_row, int mi_col,
                           int bottom_available, int left_available,
                           TX_SIZE txsz, int row_off, int col_off, int ss_y) {
  if (!bottom_available || !left_available) return 0;

  if (col_off > 0) {
    // Bottom-left pixels are in the bottom-left block, which is not available.
491
492
    return 0;
  } else {
493
494
495
    const int bh_unit = block_size_high[bsize] >> tx_size_high_log2[0];
    const int plane_bh_unit = AOMMAX(bh_unit >> ss_y, 1);
    const int bottom_left_count_unit = tx_size_high_unit[txsz];
496

497
#if !CONFIG_CB4X4
498
499
    // Special handling for block sizes 8x4 and 4x4.
    if (ss_y == 0 && bh_unit < 2 && row_off == 0) return 1;
500
#endif
501

502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
    // All bottom-left pixels are in the left block, which is already available.
    if (row_off + bottom_left_count_unit < plane_bh_unit) return 1;

    const int bw_in_mi_log2 = mi_width_log2_lookup[bsize];
    const int bh_in_mi_log2 = mi_height_log2_lookup[bsize];
    const int blk_row_in_sb = (mi_row & MAX_MIB_MASK) >> bh_in_mi_log2;
    const int blk_col_in_sb = (mi_col & MAX_MIB_MASK) >> bw_in_mi_log2;

    // Leftmost column of superblock: so bottom-left pixels maybe in the left
    // and/or bottom-left superblocks. But only the left superblock is
    // available, so check if all required pixels fall in that superblock.
    if (blk_col_in_sb == 0) {
      const int blk_start_row_off = blk_row_in_sb << (bh_in_mi_log2 + !ss_y);
      const int row_off_in_sb = blk_start_row_off + row_off;
      const int sb_height_unit = MAX_MIB_SIZE << !ss_y;
      return row_off_in_sb + bottom_left_count_unit < sb_height_unit;
    }
519

520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
    // Bottom row of superblock (and not the leftmost column): so bottom-left
    // pixels fall in the bottom superblock, which is not available yet.
    if (((blk_row_in_sb + 1) << bh_in_mi_log2) >= MAX_MIB_SIZE) return 0;

    // General case (neither leftmost column nor bottom row): check if the
    // bottom-left block is coded before the current block.
    const uint16_t *const order = orders[bsize];
    const int this_blk_index =
        ((blk_row_in_sb + 0) << (MAX_MIB_SIZE_LOG2 - bw_in_mi_log2)) +
        blk_col_in_sb + 0;
    const uint16_t this_blk_order = order[this_blk_index];
    const int bl_blk_index =
        ((blk_row_in_sb + 1) << (MAX_MIB_SIZE_LOG2 - bw_in_mi_log2)) +
        blk_col_in_sb - 1;
    const uint16_t bl_blk_order = order[bl_blk_index];
    return bl_blk_order < this_blk_order;
536
537
  }
}
Jingning Han's avatar
Jingning Han committed
538
539
540
541
542
543
544

typedef void (*intra_pred_fn)(uint8_t *dst, ptrdiff_t stride,
                              const uint8_t *above, const uint8_t *left);

static intra_pred_fn pred[INTRA_MODES][TX_SIZES];
static intra_pred_fn dc_pred[2][2][TX_SIZES];

545
#if CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
546
547
548
typedef void (*intra_high_pred_fn)(uint16_t *dst, ptrdiff_t stride,
                                   const uint16_t *above, const uint16_t *left,
                                   int bd);
549
550
static intra_high_pred_fn pred_high[INTRA_MODES][TX_SIZES];
static intra_high_pred_fn dc_pred_high[2][2][TX_SIZES];
551
#endif  // CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
552

Yaowu Xu's avatar
Yaowu Xu committed
553
static void av1_init_intra_predictors_internal(void) {
554
555
556
557
558
559
560
#if CONFIG_TX64X64
#define INIT_NO_4X4(p, type)                  \
  p[TX_8X8] = aom_##type##_predictor_8x8;     \
  p[TX_16X16] = aom_##type##_predictor_16x16; \
  p[TX_32X32] = aom_##type##_predictor_32x32; \
  p[TX_64X64] = aom_##type##_predictor_64x64
#else
clang-format's avatar
clang-format committed
561
#define INIT_NO_4X4(p, type)                  \
Yaowu Xu's avatar
Yaowu Xu committed
562
563
564
  p[TX_8X8] = aom_##type##_predictor_8x8;     \
  p[TX_16X16] = aom_##type##_predictor_16x16; \
  p[TX_32X32] = aom_##type##_predictor_32x32
565
#endif  // CONFIG_TX64X64
Jingning Han's avatar
Jingning Han committed
566

Jingning Han's avatar
Jingning Han committed
567
568
569
570
571
572
#if CONFIG_CB4X4
#define INIT_ALL_SIZES(p, type)           \
  p[TX_2X2] = aom_##type##_predictor_2x2; \
  p[TX_4X4] = aom_##type##_predictor_4x4; \
  INIT_NO_4X4(p, type)
#else
clang-format's avatar
clang-format committed
573
#define INIT_ALL_SIZES(p, type)           \
Yaowu Xu's avatar
Yaowu Xu committed
574
  p[TX_4X4] = aom_##type##_predictor_4x4; \
575
  INIT_NO_4X4(p, type)
Jingning Han's avatar
Jingning Han committed
576
#endif
577

Jingning Han's avatar
Jingning Han committed
578
579
  INIT_ALL_SIZES(pred[V_PRED], v);
  INIT_ALL_SIZES(pred[H_PRED], h);
580
581
582
  INIT_ALL_SIZES(pred[D207_PRED], d207e);
  INIT_ALL_SIZES(pred[D45_PRED], d45e);
  INIT_ALL_SIZES(pred[D63_PRED], d63e);
Jingning Han's avatar
Jingning Han committed
583
584
585
  INIT_ALL_SIZES(pred[D117_PRED], d117);
  INIT_ALL_SIZES(pred[D135_PRED], d135);
  INIT_ALL_SIZES(pred[D153_PRED], d153);
Urvang Joshi's avatar
Urvang Joshi committed
586
587
588

#if CONFIG_ALT_INTRA
  INIT_ALL_SIZES(pred[TM_PRED], paeth);
589
  INIT_ALL_SIZES(pred[SMOOTH_PRED], smooth);
Urvang Joshi's avatar
Urvang Joshi committed
590
#else
Jingning Han's avatar
Jingning Han committed
591
  INIT_ALL_SIZES(pred[TM_PRED], tm);
Urvang Joshi's avatar
Urvang Joshi committed
592
#endif  // CONFIG_ALT_INTRA
Jingning Han's avatar
Jingning Han committed
593
594
595
596
597
598

  INIT_ALL_SIZES(dc_pred[0][0], dc_128);
  INIT_ALL_SIZES(dc_pred[0][1], dc_top);
  INIT_ALL_SIZES(dc_pred[1][0], dc_left);
  INIT_ALL_SIZES(dc_pred[1][1], dc);

599
#if CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
600
601
  INIT_ALL_SIZES(pred_high[V_PRED], highbd_v);
  INIT_ALL_SIZES(pred_high[H_PRED], highbd_h);
602
603
604
  INIT_ALL_SIZES(pred_high[D207_PRED], highbd_d207e);
  INIT_ALL_SIZES(pred_high[D45_PRED], highbd_d45e);
  INIT_ALL_SIZES(pred_high[D63_PRED], highbd_d63e);
Jingning Han's avatar
Jingning Han committed
605
606
607
  INIT_ALL_SIZES(pred_high[D117_PRED], highbd_d117);
  INIT_ALL_SIZES(pred_high[D135_PRED], highbd_d135);
  INIT_ALL_SIZES(pred_high[D153_PRED], highbd_d153);
Urvang Joshi's avatar
Urvang Joshi committed
608
609
610

#if CONFIG_ALT_INTRA
  INIT_ALL_SIZES(pred_high[TM_PRED], highbd_paeth);
611
  INIT_ALL_SIZES(pred_high[SMOOTH_PRED], highbd_smooth);
Urvang Joshi's avatar
Urvang Joshi committed
612
#else
Jingning Han's avatar
Jingning Han committed
613
  INIT_ALL_SIZES(pred_high[TM_PRED], highbd_tm);
Urvang Joshi's avatar
Urvang Joshi committed
614
#endif  // CONFIG_ALT_INTRA
Jingning Han's avatar
Jingning Han committed
615
616
617
618
619

  INIT_ALL_SIZES(dc_pred_high[0][0], highbd_dc_128);
  INIT_ALL_SIZES(dc_pred_high[0][1], highbd_dc_top);
  INIT_ALL_SIZES(dc_pred_high[1][0], highbd_dc_left);
  INIT_ALL_SIZES(dc_pred_high[1][1], highbd_dc);
620
#endif  // CONFIG_HIGHBITDEPTH
Jingning Han's avatar
Jingning Han committed
621
622
623
624

#undef intra_pred_allsizes
}

hui su's avatar
hui su committed
625
#if CONFIG_EXT_INTRA
hui su's avatar
hui su committed
626
#if CONFIG_INTRA_INTERP
627
628
629
630
631
632
633
634
635
636
637
static int intra_subpel_interp(int base, int shift, const uint8_t *ref,
                               int ref_start_idx, int ref_end_idx,
                               INTRA_FILTER filter_type) {
  int val, k, idx, filter_idx = 0;
  const int16_t *filter = NULL;

  if (filter_type == INTRA_FILTER_LINEAR) {
    val = ref[base] * (256 - shift) + ref[base + 1] * shift;
    val = ROUND_POWER_OF_TWO(val, 8);
  } else {
    filter_idx = ROUND_POWER_OF_TWO(shift, 8 - SUBPEL_BITS);
Yaowu Xu's avatar
Yaowu Xu committed
638
    filter = av1_intra_filter_kernels[filter_type][filter_idx];
639
640
641
642
643

    if (filter_idx < (1 << SUBPEL_BITS)) {
      val = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k) {
        idx = base + 1 - (SUBPEL_TAPS / 2) + k;
Yaowu Xu's avatar
Yaowu Xu committed
644
        idx = AOMMAX(AOMMIN(idx, ref_end_idx), ref_start_idx);
645
646
647
648
649
650
651
652
653
654
        val += ref[idx] * filter[k];
      }
      val = ROUND_POWER_OF_TWO(val, FILTER_BITS);
    } else {
      val = ref[base + 1];
    }
  }

  return val;
}
hui su's avatar
hui su committed
655
#endif  // CONFIG_INTRA_INTERP
656

hui su's avatar
hui su committed
657
658
// Directional prediction, zone 1: 0 < angle < 90
static void dr_prediction_z1(uint8_t *dst, ptrdiff_t stride, int bs,
hui su's avatar
hui su committed
659
660
661
662
663
                             const uint8_t *above, const uint8_t *left,
#if CONFIG_INTRA_INTERP
                             INTRA_FILTER filter_type,
#endif  // CONFIG_INTRA_INTERP
                             int dx, int dy) {
hui su's avatar
hui su committed
664
  int r, c, x, base, shift, val;
hui su's avatar
hui su committed
665
666
667
668

  (void)left;
  (void)dy;
  assert(dy == 1);
hui su's avatar
hui su committed
669
  assert(dx > 0);
hui su's avatar
hui su committed
670

hui su's avatar
hui su committed
671
#if CONFIG_INTRA_INTERP
hui su's avatar
hui su committed
672
673
674
  if (filter_type != INTRA_FILTER_LINEAR) {
    const int pad_size = SUBPEL_TAPS >> 1;
    int len;
675
676
    DECLARE_ALIGNED(16, uint8_t, buf[SUBPEL_SHIFTS][MAX_SB_SIZE]);
    DECLARE_ALIGNED(16, uint8_t, src[MAX_SB_SIZE + SUBPEL_TAPS]);
hui su's avatar
hui su committed
677
678
679
680
681
682
683
684
    uint8_t flags[SUBPEL_SHIFTS];

    memset(flags, 0, SUBPEL_SHIFTS * sizeof(flags[0]));
    memset(src, above[0], pad_size * sizeof(above[0]));
    memcpy(src + pad_size, above, 2 * bs * sizeof(above[0]));
    memset(src + pad_size + 2 * bs, above[2 * bs - 1],
           pad_size * sizeof(above[0]));
    flags[0] = 1;
hui su's avatar
hui su committed
685
686
    x = dx;
    for (r = 0; r < bs; ++r, dst += stride, x += dx) {
hui su's avatar
hui su committed
687
      base = x >> 8;
hui su's avatar
hui su committed
688
689
690
691
692
693
      shift = x & 0xFF;
      shift = ROUND_POWER_OF_TWO(shift, 8 - SUBPEL_BITS);
      if (shift == SUBPEL_SHIFTS) {
        base += 1;
        shift = 0;
      }
Yaowu Xu's avatar
Yaowu Xu committed
694
      len = AOMMIN(bs, 2 * bs - 1 - base);
hui su's avatar
hui su committed
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
      if (len <= 0) {
        int i;
        for (i = r; i < bs; ++i) {
          memset(dst, above[2 * bs - 1], bs * sizeof(dst[0]));
          dst += stride;
        }
        return;
      }

      if (len <= (bs >> 1) && !flags[shift]) {
        base = x >> 8;
        shift = x & 0xFF;
        for (c = 0; c < len; ++c) {
          val = intra_subpel_interp(base, shift, above, 0, 2 * bs - 1,
                                    filter_type);
          dst[c] = clip_pixel(val);
          ++base;
        }
      } else {
        if (!flags[shift]) {
Yaowu Xu's avatar
Yaowu Xu committed
715
716
          const int16_t *filter = av1_intra_filter_kernels[filter_type][shift];
          aom_convolve8_horiz(src + pad_size, 2 * bs, buf[shift], 2 * bs,
clang-format's avatar
clang-format committed
717
718
                              filter, 16, NULL, 16, 2 * bs,
                              2 * bs < 16 ? 2 : 1);
hui su's avatar
hui su committed
719
720
721
          flags[shift] = 1;
        }
        memcpy(dst, shift == 0 ? src + pad_size + base : &buf[shift][base],
clang-format's avatar
clang-format committed
722
               len * sizeof(dst[0]));
hui su's avatar
hui su committed
723
724
725
726
727
728
729
      }

      if (len < bs)
        memset(dst + len, above[2 * bs - 1], (bs - len) * sizeof(dst[0]));
    }
    return;
  }
hui su's avatar
hui su committed
730
#endif  // CONFIG_INTRA_INTERP
hui su's avatar
hui su committed
731

hui su's avatar
hui su committed
732
733
  x = dx;
  for (r = 0; r < bs; ++r, dst += stride, x += dx) {
hui su's avatar
hui su committed
734
735
736
737
738
739
740
741
742
743
744
745
746
    base = x >> 8;
    shift = x & 0xFF;

    if (base >= 2 * bs - 1) {
      int i;
      for (i = r; i < bs; ++i) {
        memset(dst, above[2 * bs - 1], bs * sizeof(dst[0]));
        dst += stride;
      }
      return;
    }

    for (c = 0; c < bs; ++c, ++base) {
hui su's avatar
hui su committed
747
      if (base < 2 * bs - 1) {
hui su's avatar
hui su committed
748
749
        val = above[base] * (256 - shift) + above[base + 1] * shift;
        val = ROUND_POWER_OF_TWO(val, 8);
hui su's avatar
hui su committed
750
751
752
753
754
755
756
757
758
759
        dst[c] = clip_pixel(val);
      } else {
        dst[c] = above[2 * bs - 1];
      }
    }
  }
}

// Directional prediction, zone 2: 90 < angle < 180
static void dr_prediction_z2(uint8_t *dst, ptrdiff_t stride, int bs,
hui su's avatar
hui su committed
760
761
762
763
764
                             const uint8_t *above, const uint8_t *left,
#if CONFIG_INTRA_INTERP
                             INTRA_FILTER filter_type,
#endif  // CONFIG_INTRA_INTERP
                             int dx, int dy) {
Debargha Mukherjee's avatar
Debargha Mukherjee committed
765
  int r, c, x, y, shift1, shift2, val, base1, base2;
hui su's avatar
hui su committed
766
767
768
769

  assert(dx > 0);
  assert(dy > 0);

hui su's avatar
hui su committed
770
771
772
773
774
775
776
  x = -dx;
  for (r = 0; r < bs; ++r, x -= dx, dst += stride) {
    base1 = x >> 8;
    y = (r << 8) - dy;
    for (c = 0; c < bs; ++c, ++base1, y -= dy) {
      if (base1 >= -1) {
        shift1 = x & 0xFF;
hui su's avatar
hui su committed
777
#if CONFIG_INTRA_INTERP
clang-format's avatar
clang-format committed
778
779
        val =
            intra_subpel_interp(base1, shift1, above, -1, bs - 1, filter_type);
hui su's avatar
hui su committed
780
781
782
783
#else
        val = above[base1] * (256 - shift1) + above[base1 + 1] * shift1;
        val = ROUND_POWER_OF_TWO(val, 8);
#endif  // CONFIG_INTRA_INTERP
hui su's avatar
hui su committed
784
      } else {
hui su's avatar
hui su committed
785
        base2 = y >> 8;
786
        shift2 = y & 0xFF;
hui su's avatar
hui su committed
787
#if CONFIG_INTRA_INTERP
788
        val = intra_subpel_interp(base2, shift2, left, -1, bs - 1, filter_type);
hui su's avatar
hui su committed
789
#else
790
791
        val = left[base2] * (256 - shift2) + left[base2 + 1] * shift2;
        val = ROUND_POWER_OF_TWO(val, 8);
hui su's avatar
hui su committed
792
#endif  // CONFIG_INTRA_INTERP
hui su's avatar
hui su committed
793
794
795
796
797
798
799
800
      }
      dst[c] = clip_pixel(val);
    }
  }
}

// Directional prediction, zone 3: 180 < angle < 270
static void dr_prediction_z3(uint8_t *dst, ptrdiff_t stride, int bs,
hui su's avatar
hui su committed
801
802
803
804
805
                             const uint8_t *above, const uint8_t *left,
#if CONFIG_INTRA_INTERP
                             INTRA_FILTER filter_type,
#endif  // CONFIG_INTRA_INTERP
                             int dx, int dy) {
hui su's avatar
hui su committed
806
  int r, c, y, base, shift, val;
hui su's avatar
hui su committed
807
808
809

  (void)above;
  (void)dx;
810

hui su's avatar
hui su committed
811
  assert(dx == 1);
hui su's avatar
hui su committed
812
  assert(dy > 0);
hui su's avatar
hui su committed
813

hui su's avatar
hui su committed
814
#if CONFIG_INTRA_INTERP
hui su's avatar
hui su committed
815
816
817
  if (filter_type != INTRA_FILTER_LINEAR) {
    const int pad_size = SUBPEL_TAPS >> 1;
    int len, i;
818
819
    DECLARE_ALIGNED(16, uint8_t, buf[MAX_SB_SIZE][4 * SUBPEL_SHIFTS]);
    DECLARE_ALIGNED(16, uint8_t, src[(MAX_SB_SIZE + SUBPEL_TAPS) * 4]);
hui su's avatar
hui su committed
820
821
822
    uint8_t flags[SUBPEL_SHIFTS];

    memset(flags, 0, SUBPEL_SHIFTS * sizeof(flags[0]));
clang-format's avatar
clang-format committed
823
824
    for (i = 0; i < pad_size; ++i) src[4 * i] = left[0];
    for (i = 0; i < 2 * bs; ++i) src[4 * (i + pad_size)] = left[i];
hui su's avatar
hui su committed
825
826
827
    for (i = 0; i < pad_size; ++i)
      src[4 * (i + 2 * bs + pad_size)] = left[2 * bs - 1];
    flags[0] = 1;
hui su's avatar
hui su committed
828
829
    y = dy;
    for (c = 0; c < bs; ++c, y += dy) {
hui su's avatar
hui su committed
830
      base = y >> 8;
hui su's avatar
hui su committed
831
832
833
834
835
836
      shift = y & 0xFF;
      shift = ROUND_POWER_OF_TWO(shift, 8 - SUBPEL_BITS);
      if (shift == SUBPEL_SHIFTS) {
        base += 1;
        shift = 0;
      }
Yaowu Xu's avatar
Yaowu Xu committed
837
      len = AOMMIN(bs, 2 * bs - 1 - base);
hui su's avatar
hui su committed
838
839

      if (len <= 0) {
Debargha Mukherjee's avatar
Debargha Mukherjee committed
840
        for (r = 0; r < bs; ++r) {
clang-format's avatar
clang-format committed
841
          dst[r * stride + c] = left[2 * bs - 1];
hui su's avatar
hui su committed
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
        }
        continue;
      }

      if (len <= (bs >> 1) && !flags[shift]) {
        base = y >> 8;
        shift = y & 0xFF;
        for (r = 0; r < len; ++r) {
          val = intra_subpel_interp(base, shift, left, 0, 2 * bs - 1,
                                    filter_type);
          dst[r * stride + c] = clip_pixel(val);
          ++base;
        }
      } else {
        if (!flags[shift]) {
Yaowu Xu's avatar
Yaowu Xu committed
857
858
          const int16_t *filter = av1_intra_filter_kernels[filter_type][shift];
          aom_convolve8_vert(src + 4 * pad_size, 4, buf[0] + 4 * shift,
clang-format's avatar
clang-format committed
859
                             4 * SUBPEL_SHIFTS, NULL, 16, filter, 16,
hui su's avatar
hui su committed
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
                             2 * bs < 16 ? 4 : 4, 2 * bs);
          flags[shift] = 1;
        }

        if (shift == 0) {
          for (r = 0; r < len; ++r) {
            dst[r * stride + c] = left[r + base];
          }
        } else {
          for (r = 0; r < len; ++r) {
            dst[r * stride + c] = buf[r + base][4 * shift];
          }
        }
      }

      if (len < bs) {
        for (r = len; r < bs; ++r) {
clang-format's avatar
clang-format committed
877
          dst[r * stride + c] = left[2 * bs - 1];
hui su's avatar
hui su committed
878
879
880
881
882
        }
      }
    }
    return;
  }
hui su's avatar
hui su committed
883
#endif  // CONFIG_INTRA_INTERP
hui su's avatar
hui su committed
884

hui su's avatar
hui su committed
885
886
  y = dy;
  for (c = 0; c < bs; ++c, y += dy) {
hui su's avatar
hui su committed
887
888
889
890
    base = y >> 8;
    shift = y & 0xFF;

    for (r = 0; r < bs; ++r, ++base) {
891
      if (base < 2 * bs - 1) {
hui su's avatar
hui su committed
892
893
894
        val = left[base] * (256 - shift) + left[base + 1] * shift;
        val = ROUND_POWER_OF_TWO(val, 8);
        dst[r * stride + c] = clip_pixel(val);
hui su's avatar
hui su committed
895
      } else {
clang-format's avatar
clang-format committed
896
        for (; r < bs; ++r) dst[r * stride + c] = left[2 * bs - 1];
hui su's avatar
hui su committed
897
        break;
hui su's avatar
hui su committed
898
899
900
901
902
      }
    }
  }
}

903
904
905
906
// Get the shift (up-scaled by 256) in X w.r.t a unit change in Y.
// If angle > 0 && angle < 90, dx = -((int)(256 / t));
// If angle > 90 && angle < 180, dx = (int)(256 / t);
// If angle > 180 && angle < 270, dx = 1;
Yaowu Xu's avatar
Yaowu Xu committed
907
static INLINE int get_dx(int angle) {
908
  if (angle > 0 && angle < 90) {
hui su's avatar
hui su committed
909
    return dr_intra_derivative[angle];
910
911
912
913
914
915
916
917
918
919
920
921
  } else if (angle > 90 && angle < 180) {
    return dr_intra_derivative[180 - angle];
  } else {
    // In this case, we are not really going to use dx. We may return any value.
    return 1;
  }
}

// Get the shift (up-scaled by 256) in Y w.r.t a unit change in X.
// If angle > 0 && angle < 90, dy = 1;
// If angle > 90 && angle < 180, dy = (int)(256 * t);
// If angle > 180 && angle < 270, dy = -((int)(256 * t));
Yaowu Xu's avatar
Yaowu Xu committed
922
static INLINE int get_dy(int angle) {
923
  if (angle > 90 && angle < 180) {
clang-format's avatar
clang-format committed
924
925
    return dr_intra_derivative[angle - 90];
  } else if (angle > 180 && angle < 270) {
hui su's avatar
hui su committed
926
    return dr_intra_derivative[270 - angle];
clang-format's avatar
clang-format committed
927
  } else {
928
929
930
931
932
    // In this case, we are not really going to use dy. We may return any value.
    return 1;
  }
}

hui su's avatar
hui su committed
933
static void dr_predictor(uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size,
hui su's avatar
hui su committed
934
935
936
937
938
                         const uint8_t *above, const uint8_t *left,
#if CONFIG_INTRA_INTERP
                         INTRA_FILTER filter_type,
#endif  // CONFIG_INTRA_INTERP
                         int angle) {
939
940
  const int dx = get_dx(angle);
  const int dy = get_dy(angle);
941
  const int bs = tx_size_wide[tx_size];
942
943
  assert(angle > 0 && angle < 270);

hui su's avatar
hui su committed
944
  if (angle > 0 && angle < 90) {
hui su's avatar
hui su committed
945
946
947
948
949
    dr_prediction_z1(dst, stride, bs, above, left,
#if CONFIG_INTRA_INTERP
                     filter_type,
#endif  // CONFIG_INTRA_INTERP
                     dx, dy);
hui su's avatar
hui su committed
950
  } else if (angle > 90 && angle < 180) {
hui su's avatar
hui su committed
951
952
953
954
955
    dr_prediction_z2(dst, stride, bs, above, left,
#if CONFIG_INTRA_INTERP
                     filter_type,
#endif  // CONFIG_INTRA_INTERP
                     dx, dy);
hui su's avatar
hui su committed
956
  } else if (angle > 180 && angle < 270) {
hui su's avatar
hui su committed
957
958
959
960
961
    dr_prediction_z3(dst, stride, bs, above, left,
#if CONFIG_INTRA_INTERP
                     filter_type,
#endif  // CONFIG_INTRA_INTERP
                     dx, dy);
hui su's avatar
hui su committed
962
  } else if (angle == 90) {
hui su's avatar
hui su committed
963
    pred[V_PRED][tx_size](dst, stride, above, left);
hui su's avatar
hui su committed
964
  } else if (angle == 180) {
hui su's avatar
hui su committed
965
    pred[H_PRED][tx_size](dst, stride, above, left);
hui su's avatar
hui su committed
966
967
968
  }
}

969
#if CONFIG_HIGHBITDEPTH
hui su's avatar
hui su committed
970
#if CONFIG_INTRA_INTERP
971
972
973
974
975
976
977
978
979
980
981
static int highbd_intra_subpel_interp(int base, int shift, const uint16_t *ref,
                                      int ref_start_idx, int ref_end_idx,
                                      INTRA_FILTER filter_type) {
  int val, k, idx, filter_idx = 0;
  const int16_t *filter = NULL;

  if (filter_type == INTRA_FILTER_LINEAR) {
    val = ref[base] * (256 - shift) + ref[base + 1] * shift;
    val = ROUND_POWER_OF_TWO(val, 8);
  } else {
    filter_idx = ROUND_POWER_OF_TWO(shift, 8 - SUBPEL_BITS);
Yaowu Xu's avatar
Yaowu Xu committed
982
    filter = av1_intra_filter_kernels[filter_type][filter_idx];
983
984
985
986
987

    if (filter_idx < (1 << SUBPEL_BITS)) {
      val = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k) {
        idx = base + 1 - (SUBPEL_TAPS / 2) + k;
Yaowu Xu's avatar
Yaowu Xu committed
988
        idx = AOMMAX(AOMMIN(idx, ref_end_idx), ref_start_idx);
989
990
991
992
993
994
995
996
997
998
        val += ref[idx] * filter[k];
      }
      val = ROUND_POWER_OF_TWO(val, FILTER_BITS);
    } else {
      val = ref[base + 1];
    }
  }

  return val;
}
hui su's avatar
hui su committed
999
#endif  // CONFIG_INTRA_INTERP
1000

hui su's avatar
hui su committed
1001
1002
1003
// Directional prediction, zone 1: 0 < angle < 90
static void highbd_dr_prediction_z1(uint16_t *dst, ptrdiff_t stride, int bs,
                                    const uint16_t *above, const uint16_t *left,
hui su's avatar
hui su committed
1004
1005
1006
1007
#if CONFIG_INTRA_INTERP
                                    INTRA_FILTER filter_type,
#endif  // CONFIG_INTRA_INTERP
                                    int dx, int dy, int bd) {
1008
  int r, c, x, base, shift, val;
hui su's avatar
hui su committed
1009
1010
1011
1012

  (void)left;
  (void)dy;
  assert(dy == 1);
1013
  assert(dx > 0);
hui su's avatar
hui su committed
1014

1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
  x = dx;
  for (r = 0; r < bs; ++r, dst += stride, x += dx) {
    base = x >> 8;
    shift = x & 0xFF;

    if (base >= 2 * bs - 1) {
      int i;
      for (i = r; i < bs; ++i) {
        aom_memset16(dst, above[2 * bs - 1], bs);
        dst += stride;
      }
      return;
    }

    for (c = 0; c < bs; ++c, ++base) {
hui su's avatar
hui su committed
1030
      if (base < 2 * bs - 1) {
hui su's avatar
hui su committed
1031
#if CONFIG_INTRA_INTERP
1032
1033
        val = highbd_intra_subpel_interp(base, shift, above, 0, 2 * bs - 1,
                                         filter_type);
hui su's avatar
hui su committed
1034
1035
1036
1037
#else
        val = above[base] * (256 - shift) + above[base + 1] * shift;
        val = ROUND_POWER_OF_TWO(val, 8);
#endif  // CONFIG_INTRA_INTERP
hui su's avatar
hui su committed
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
        dst[c] = clip_pixel_highbd(val, bd);
      } else {
        dst[c] = above[2 * bs - 1];
      }
    }
  }
}

// Directional prediction, zone 2: 90 < angle < 180
static void highbd_dr_prediction_z2(uint16_t *dst, ptrdiff_t stride, int bs,
                                    const uint16_t *above, const uint16_t *left,
hui su's avatar
hui su committed
1049
1050
1051
1052
#if CONFIG_INTRA_INTERP
                                    INTRA_FILTER filter_type,
#endif  // CONFIG_INTRA_INTERP
                                    int dx, int dy, int bd) {
1053
  int r, c, x, y, shift, val, base;
hui su's avatar
hui su committed
1054
1055
1056
1057
1058
1059
1060

  assert(dx > 0);
  assert(dy > 0);

  for (r = 0; r < bs; ++r) {
    for (c = 0; c < bs; ++c) {
      y = r + 1;
1061
1062
1063
      x = (c << 8) - y * dx;
      base = x >> 8;
      if (base >= -1) {
1064
        shift = x & 0xFF;
hui su's avatar
hui su committed
1065
#if CONFIG_INTRA_INTERP
1066
1067
        val = highbd_intra_subpel_interp(base, shift, above, -1, bs - 1,
                                         filter_type);
hui su's avatar
hui su committed
1068
1069
1070
1071
#else
        val = above[base] * (256 - shift) + above[base + 1] * shift;
        val = ROUND_POWER_OF_TWO(val, 8);
#endif  // CONFIG_INTRA_INTERP
hui su's avatar
hui su committed
1072
1073
      } else {
        x = c + 1;